Skip to content

~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.

7.2 KB
Raw Blame
"""High-value tests for utilities/json_utils.py pure logic."""

import json
import unittest
from unittest.mock import MagicMock

from local_deep_research.utilities.json_utils import (
    extract_json,
    get_llm_response_text,
    _clean_llm_json_artifacts,
    _extract_by_brackets,
    _remove_think_tags,
    _strip_code_fences,
)


class TestGetLlmResponseText(unittest.TestCase):
    def test_none_returns_empty_string(self):
        assert get_llm_response_text(None) == ""

    def test_content_attribute(self):
        msg = MagicMock()
        msg.content = "hello world"
        assert get_llm_response_text(msg) == "hello world"

    def test_text_attribute_when_no_content(self):
        msg = MagicMock(spec=[])
        msg.text = "from text"
        assert get_llm_response_text(msg) == "from text"

    def test_str_fallback(self):
        assert get_llm_response_text(42) == "42"

    def test_think_tags_removed(self):
        msg = MagicMock()
        msg.content = "before <think>inner</think> after"
        assert get_llm_response_text(msg) == "before  after"

    def test_content_none_falls_to_text(self):
        msg = MagicMock()
        msg.content = None
        msg.text = "fallback"
        assert get_llm_response_text(msg) == "fallback"

    def test_non_string_content_converted(self):
        msg = MagicMock()
        msg.content = 123
        result = get_llm_response_text(msg)
        assert result == "123"


class TestRemoveThinkTags(unittest.TestCase):
    def test_paired_tags(self):
        assert _remove_think_tags("a <think>b</think> c") == "a  c"

    def test_orphaned_closing(self):
        assert _remove_think_tags("text </think> more") == "text  more"

    def test_orphaned_opening(self):
        assert _remove_think_tags("text <think> more") == "text  more"

    def test_multiline_think(self):
        text = "start <think>\nline1\nline2\n</think> end"
        assert _remove_think_tags(text) == "start  end"

    def test_no_tags(self):
        assert _remove_think_tags("plain text") == "plain text"

    def test_empty_string(self):
        assert _remove_think_tags("") == ""


class TestStripCodeFences(unittest.TestCase):
    def test_json_fence(self):
        text = '```json\n{"key": "val"}\n```'
        assert _strip_code_fences(text) == '{"key": "val"}'

    def test_plain_fence(self):
        text = '```\n{"key": "val"}\n```'
        assert _strip_code_fences(text) == '{"key": "val"}'

    def test_no_fence(self):
        text = '{"key": "val"}'
        assert _strip_code_fences(text) == text

    def test_json_fence_with_surrounding_prose(self):
        text = 'Here is the result:\n```json\n{"a": 1}\n```\nDone.'
        assert _strip_code_fences(text) == '{"a": 1}'

    def test_unclosed_fence_returns_original(self):
        text = "```only one fence"
        assert _strip_code_fences(text) == text


class TestExtractByBrackets(unittest.TestCase):
    def test_curly_brackets(self):
        text = 'prefix {"a": 1} suffix'
        assert _extract_by_brackets(text, "{", "}") == '{"a": 1}'

    def test_square_brackets(self):
        text = "before [1, 2, 3] after"
        assert _extract_by_brackets(text, "[", "]") == "[1, 2, 3]"

    def test_nested_brackets(self):
        text = '{"a": {"b": 1}}'
        assert _extract_by_brackets(text, "{", "}") == '{"a": {"b": 1}}'

    def test_no_brackets_returns_none(self):
        assert _extract_by_brackets("no brackets", "{", "}") is None

    def test_mismatched_order_returns_none(self):
        assert _extract_by_brackets("} before {", "{", "}") is None

    def test_only_open_returns_none(self):
        assert _extract_by_brackets("{ no close", "{", "}") is None


class TestCleanLlmJsonArtifacts(unittest.TestCase):
    def test_trailing_comma_before_brace(self):
        text = '{"a": 1,}'
        assert _clean_llm_json_artifacts(text) == '{"a": 1}'

    def test_trailing_comma_before_bracket(self):
        text = "[1, 2, 3,]"
        assert _clean_llm_json_artifacts(text) == "[1, 2, 3]"

    def test_line_comments(self):
        text = '{"a": 1 // comment\n}'
        result = _clean_llm_json_artifacts(text)
        assert "//" not in result

    def test_ellipsis_entries(self):
        text = '["a", ..., "b"]'
        result = _clean_llm_json_artifacts(text)
        assert "..." not in result
        parsed = json.loads(result)
        assert parsed == ["a", "b"]

    def test_quoted_ellipsis(self):
        text = '["a", "...", "b"]'
        result = _clean_llm_json_artifacts(text)
        assert "..." not in result
        parsed = json.loads(result)
        assert parsed == ["a", "b"]

    def test_no_artifacts_unchanged(self):
        text = '{"a": 1}'
        assert _clean_llm_json_artifacts(text) == text


class TestExtractJsonDirect(unittest.TestCase):
    def test_none_returns_none(self):
        assert extract_json(None) is None

    def test_empty_string_returns_none(self):
        assert extract_json("") is None

    def test_whitespace_only_returns_none(self):
        assert extract_json("   ") is None

    def test_valid_dict(self):
        assert extract_json('{"key": "val"}') == {"key": "val"}

    def test_valid_list(self):
        assert extract_json("[1, 2, 3]") == [1, 2, 3]

    def test_invalid_json_returns_none(self):
        assert extract_json("not json at all") is None


class TestExtractJsonExpectedType(unittest.TestCase):
    def test_expect_dict_gets_dict(self):
        result = extract_json('{"a": 1}', expected_type=dict)
        assert result == {"a": 1}

    def test_expect_list_gets_list(self):
        result = extract_json("[1, 2]", expected_type=list)
        assert result == [1, 2]

    def test_expect_dict_skips_list(self):
        # Text has both a list and a dict
        text = 'Here is [1] and {"a": 1}'
        result = extract_json(text, expected_type=dict)
        assert isinstance(result, dict)

    def test_expect_list_skips_dict(self):
        text = 'Here is {"a": 1} and [1, 2]'
        result = extract_json(text, expected_type=list)
        assert isinstance(result, list)

    def test_type_mismatch_falls_through(self):
        # Only a list in text, but dict expected — returns None
        result = extract_json("[1, 2]", expected_type=dict)
        assert result is None


class TestExtractJsonComplex(unittest.TestCase):
    def test_json_in_code_fence(self):
        text = '```json\n{"a": 1}\n```'
        assert extract_json(text) == {"a": 1}

    def test_json_with_surrounding_prose(self):
        text = 'The answer is: {"result": true} as shown.'
        assert extract_json(text) == {"result": True}

    def test_json_with_think_tags(self):
        text = '<think>reasoning</think>{"answer": 42}'
        assert extract_json(text) == {"answer": 42}

    def test_json_with_trailing_comma(self):
        text = '{"a": 1, "b": 2,}'
        result = extract_json(text)
        assert result == {"a": 1, "b": 2}

    def test_json_with_comments(self):
        text = '{"a": 1 // this is a comment\n}'
        result = extract_json(text)
        assert result == {"a": 1}

    def test_nested_json(self):
        text = '{"outer": {"inner": [1, 2, 3]}}'
        result = extract_json(text)
        assert result == {"outer": {"inner": [1, 2, 3]}}


if __name__ == "__main__":
    unittest.main()